The American artificial intelligence (AI) start-up Anthropic has discovered an operation linked to the United Arab Emirates that notably targeted UN experts critical of Abu Dhabi's alleged role in the conflict, according to a document published Thursday evening.

In a report explaining how it stopped illegal operations from several countries, Anthropic speaks of a large-scale operation, which it attributes "with a high degree of certainty" to the Emirati government.

Primarily targeting the Muslim Brotherhood, this campaign used Anthropic's AI assistant agents, Claude, to generate various documents, content published on social networks, and testimonies about Sudan.

The war, which began in 2023, between the Sudanese army and the paramilitaries of the Rapid Support Forces (RSF) has killed more than 200,000 people, according to humanitarian workers, and displaced more than 12 million Sudanese, in what the UN describes as the world's worst humanitarian crisis.

The United Arab Emirates, which categorically denies the allegations, have been repeatedly accused of supporting the FSR and fueling the conflict.

According to Anthropic, the operation revolved around a virtual persona named "Deadshot" calling for "a coordinated regional and transatlantic operation to dismantle the Muslim Brotherhood worldwide."

Founded in 1928 in Egypt, this Sunni pan-Islamist movement has been suspected of links with the Sudanese army, which the latter denies.

Another initiative involved appropriating the identity of a Sudanese NGO and writing the testimonies of two individuals so that "elements favorable to the interests of one of the two sides of the conflict would reach the UN as if they came from independent local witnesses, rather than from state actors," Anthropic continues.

The American group also points to the creation of dossiers aimed at undermining UN rapporteurs who had criticized Abu Dhabi's role in Sudan.

When contacted by AFP, the Emirates did not immediately respond.

The operation exploited approximately 300 fake social media accounts, according to Anthropic, which was nevertheless unable to determine whether the testimonies or files reached their target or influenced political decisions.

The start-up announced on Thursday that it had stopped several surveillance operations carried out against dissidents and ethnic minorities, notably in China, Iran and West Africa, but also attempts to design weapons, such as in Yemen, or even large-scale scams.

"It is high time that these suppliers put their affairs in order and ensure that they keep their advanced models under control," European Commission spokesman Thomas Regnier told reporters on Friday.